1 //=-- InstrProfReader.h - Instrumented profiling readers ----------*- C++ -*-=// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This file contains support for reading profiling data for instrumentation 11 // based PGO and coverage. 12 // 13 //===----------------------------------------------------------------------===// 14 15 #ifndef LLVM_PROFILEDATA_INSTRPROFREADER_H 16 #define LLVM_PROFILEDATA_INSTRPROFREADER_H 17 18 #include "llvm/ADT/ArrayRef.h" 19 #include "llvm/ADT/StringExtras.h" 20 #include "llvm/ProfileData/InstrProf.h" 21 #include "llvm/Support/EndianStream.h" 22 #include "llvm/Support/ErrorOr.h" 23 #include "llvm/Support/LineIterator.h" 24 #include "llvm/Support/MemoryBuffer.h" 25 #include "llvm/Support/OnDiskHashTable.h" 26 #include <iterator> 27 28 namespace llvm { 29 30 class InstrProfReader; 31 32 /// Profiling information for a single function. 33 struct InstrProfRecord { InstrProfRecordInstrProfRecord34 InstrProfRecord() {} InstrProfRecordInstrProfRecord35 InstrProfRecord(StringRef Name, uint64_t Hash, ArrayRef<uint64_t> Counts) 36 : Name(Name), Hash(Hash), Counts(Counts) {} 37 StringRef Name; 38 uint64_t Hash; 39 ArrayRef<uint64_t> Counts; 40 }; 41 42 /// A file format agnostic iterator over profiling data. 43 class InstrProfIterator : public std::iterator<std::input_iterator_tag, 44 InstrProfRecord> { 45 InstrProfReader *Reader; 46 InstrProfRecord Record; 47 48 void Increment(); 49 public: InstrProfIterator()50 InstrProfIterator() : Reader(nullptr) {} InstrProfIterator(InstrProfReader * Reader)51 InstrProfIterator(InstrProfReader *Reader) : Reader(Reader) { Increment(); } 52 53 InstrProfIterator &operator++() { Increment(); return *this; } 54 bool operator==(const InstrProfIterator &RHS) { return Reader == RHS.Reader; } 55 bool operator!=(const InstrProfIterator &RHS) { return Reader != RHS.Reader; } 56 InstrProfRecord &operator*() { return Record; } 57 InstrProfRecord *operator->() { return &Record; } 58 }; 59 60 /// Base class and interface for reading profiling data of any known instrprof 61 /// format. Provides an iterator over InstrProfRecords. 62 class InstrProfReader { 63 std::error_code LastError; 64 65 public: InstrProfReader()66 InstrProfReader() : LastError(instrprof_error::success) {} ~InstrProfReader()67 virtual ~InstrProfReader() {} 68 69 /// Read the header. Required before reading first record. 70 virtual std::error_code readHeader() = 0; 71 /// Read a single record. 72 virtual std::error_code readNextRecord(InstrProfRecord &Record) = 0; 73 /// Iterator over profile data. begin()74 InstrProfIterator begin() { return InstrProfIterator(this); } end()75 InstrProfIterator end() { return InstrProfIterator(); } 76 77 protected: 78 /// Set the current std::error_code and return same. error(std::error_code EC)79 std::error_code error(std::error_code EC) { 80 LastError = EC; 81 return EC; 82 } 83 84 /// Clear the current error code and return a successful one. success()85 std::error_code success() { return error(instrprof_error::success); } 86 87 public: 88 /// Return true if the reader has finished reading the profile data. isEOF()89 bool isEOF() { return LastError == instrprof_error::eof; } 90 /// Return true if the reader encountered an error reading profiling data. hasError()91 bool hasError() { return LastError && !isEOF(); } 92 /// Get the current error code. getError()93 std::error_code getError() { return LastError; } 94 95 /// Factory method to create an appropriately typed reader for the given 96 /// instrprof file. 97 static ErrorOr<std::unique_ptr<InstrProfReader>> create(std::string Path); 98 }; 99 100 /// Reader for the simple text based instrprof format. 101 /// 102 /// This format is a simple text format that's suitable for test data. Records 103 /// are separated by one or more blank lines, and record fields are separated by 104 /// new lines. 105 /// 106 /// Each record consists of a function name, a function hash, a number of 107 /// counters, and then each counter value, in that order. 108 class TextInstrProfReader : public InstrProfReader { 109 private: 110 /// The profile data file contents. 111 std::unique_ptr<MemoryBuffer> DataBuffer; 112 /// Iterator over the profile data. 113 line_iterator Line; 114 /// The current set of counter values. 115 std::vector<uint64_t> Counts; 116 117 TextInstrProfReader(const TextInstrProfReader &) LLVM_DELETED_FUNCTION; 118 TextInstrProfReader &operator=(const TextInstrProfReader &) 119 LLVM_DELETED_FUNCTION; 120 public: TextInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer_)121 TextInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer_) 122 : DataBuffer(std::move(DataBuffer_)), Line(*DataBuffer, true, '#') {} 123 124 /// Read the header. readHeader()125 std::error_code readHeader() override { return success(); } 126 /// Read a single record. 127 std::error_code readNextRecord(InstrProfRecord &Record) override; 128 }; 129 130 /// Reader for the raw instrprof binary format from runtime. 131 /// 132 /// This format is a raw memory dump of the instrumentation-baed profiling data 133 /// from the runtime. It has no index. 134 /// 135 /// Templated on the unsigned type whose size matches pointers on the platform 136 /// that wrote the profile. 137 template <class IntPtrT> 138 class RawInstrProfReader : public InstrProfReader { 139 private: 140 /// The profile data file contents. 141 std::unique_ptr<MemoryBuffer> DataBuffer; 142 /// The current set of counter values. 143 std::vector<uint64_t> Counts; 144 struct ProfileData { 145 const uint32_t NameSize; 146 const uint32_t NumCounters; 147 const uint64_t FuncHash; 148 const IntPtrT NamePtr; 149 const IntPtrT CounterPtr; 150 }; 151 struct RawHeader { 152 const uint64_t Magic; 153 const uint64_t Version; 154 const uint64_t DataSize; 155 const uint64_t CountersSize; 156 const uint64_t NamesSize; 157 const uint64_t CountersDelta; 158 const uint64_t NamesDelta; 159 }; 160 161 bool ShouldSwapBytes; 162 uint64_t CountersDelta; 163 uint64_t NamesDelta; 164 const ProfileData *Data; 165 const ProfileData *DataEnd; 166 const uint64_t *CountersStart; 167 const char *NamesStart; 168 const char *ProfileEnd; 169 170 RawInstrProfReader(const RawInstrProfReader &) LLVM_DELETED_FUNCTION; 171 RawInstrProfReader &operator=(const RawInstrProfReader &) 172 LLVM_DELETED_FUNCTION; 173 public: RawInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer)174 RawInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer) 175 : DataBuffer(std::move(DataBuffer)) { } 176 177 static bool hasFormat(const MemoryBuffer &DataBuffer); 178 std::error_code readHeader() override; 179 std::error_code readNextRecord(InstrProfRecord &Record) override; 180 181 private: 182 std::error_code readNextHeader(const char *CurrentPos); 183 std::error_code readHeader(const RawHeader &Header); 184 template <class IntT> swap(IntT Int)185 IntT swap(IntT Int) const { 186 return ShouldSwapBytes ? sys::getSwappedBytes(Int) : Int; 187 } getCounter(IntPtrT CounterPtr)188 const uint64_t *getCounter(IntPtrT CounterPtr) const { 189 ptrdiff_t Offset = (swap(CounterPtr) - CountersDelta) / sizeof(uint64_t); 190 return CountersStart + Offset; 191 } getName(IntPtrT NamePtr)192 const char *getName(IntPtrT NamePtr) const { 193 ptrdiff_t Offset = (swap(NamePtr) - NamesDelta) / sizeof(char); 194 return NamesStart + Offset; 195 } 196 }; 197 198 typedef RawInstrProfReader<uint32_t> RawInstrProfReader32; 199 typedef RawInstrProfReader<uint64_t> RawInstrProfReader64; 200 201 namespace IndexedInstrProf { 202 enum class HashT : uint32_t; 203 } 204 205 /// Trait for lookups into the on-disk hash table for the binary instrprof 206 /// format. 207 class InstrProfLookupTrait { 208 std::vector<uint64_t> DataBuffer; 209 IndexedInstrProf::HashT HashType; 210 public: InstrProfLookupTrait(IndexedInstrProf::HashT HashType)211 InstrProfLookupTrait(IndexedInstrProf::HashT HashType) : HashType(HashType) {} 212 213 struct data_type { data_typedata_type214 data_type(StringRef Name, ArrayRef<uint64_t> Data) 215 : Name(Name), Data(Data) {} 216 StringRef Name; 217 ArrayRef<uint64_t> Data; 218 }; 219 typedef StringRef internal_key_type; 220 typedef StringRef external_key_type; 221 typedef uint64_t hash_value_type; 222 typedef uint64_t offset_type; 223 EqualKey(StringRef A,StringRef B)224 static bool EqualKey(StringRef A, StringRef B) { return A == B; } GetInternalKey(StringRef K)225 static StringRef GetInternalKey(StringRef K) { return K; } 226 227 hash_value_type ComputeHash(StringRef K); 228 229 static std::pair<offset_type, offset_type> ReadKeyDataLength(const unsigned char * & D)230 ReadKeyDataLength(const unsigned char *&D) { 231 using namespace support; 232 offset_type KeyLen = endian::readNext<offset_type, little, unaligned>(D); 233 offset_type DataLen = endian::readNext<offset_type, little, unaligned>(D); 234 return std::make_pair(KeyLen, DataLen); 235 } 236 ReadKey(const unsigned char * D,offset_type N)237 StringRef ReadKey(const unsigned char *D, offset_type N) { 238 return StringRef((const char *)D, N); 239 } 240 ReadData(StringRef K,const unsigned char * D,offset_type N)241 data_type ReadData(StringRef K, const unsigned char *D, offset_type N) { 242 DataBuffer.clear(); 243 if (N % sizeof(uint64_t)) 244 // The data is corrupt, don't try to read it. 245 return data_type("", DataBuffer); 246 247 using namespace support; 248 // We just treat the data as opaque here. It's simpler to handle in 249 // IndexedInstrProfReader. 250 unsigned NumEntries = N / sizeof(uint64_t); 251 DataBuffer.reserve(NumEntries); 252 for (unsigned I = 0; I < NumEntries; ++I) 253 DataBuffer.push_back(endian::readNext<uint64_t, little, unaligned>(D)); 254 return data_type(K, DataBuffer); 255 } 256 }; 257 typedef OnDiskIterableChainedHashTable<InstrProfLookupTrait> 258 InstrProfReaderIndex; 259 260 /// Reader for the indexed binary instrprof format. 261 class IndexedInstrProfReader : public InstrProfReader { 262 private: 263 /// The profile data file contents. 264 std::unique_ptr<MemoryBuffer> DataBuffer; 265 /// The index into the profile data. 266 std::unique_ptr<InstrProfReaderIndex> Index; 267 /// Iterator over the profile data. 268 InstrProfReaderIndex::data_iterator RecordIterator; 269 /// Offset into our current data set. 270 size_t CurrentOffset; 271 /// The file format version of the profile data. 272 uint64_t FormatVersion; 273 /// The maximal execution count among all functions. 274 uint64_t MaxFunctionCount; 275 276 IndexedInstrProfReader(const IndexedInstrProfReader &) LLVM_DELETED_FUNCTION; 277 IndexedInstrProfReader &operator=(const IndexedInstrProfReader &) 278 LLVM_DELETED_FUNCTION; 279 public: IndexedInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer)280 IndexedInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer) 281 : DataBuffer(std::move(DataBuffer)), Index(nullptr), CurrentOffset(0) {} 282 283 /// Return true if the given buffer is in an indexed instrprof format. 284 static bool hasFormat(const MemoryBuffer &DataBuffer); 285 286 /// Read the file header. 287 std::error_code readHeader() override; 288 /// Read a single record. 289 std::error_code readNextRecord(InstrProfRecord &Record) override; 290 291 /// Fill Counts with the profile data for the given function name. 292 std::error_code getFunctionCounts(StringRef FuncName, uint64_t FuncHash, 293 std::vector<uint64_t> &Counts); 294 /// Return the maximum of all known function counts. getMaximumFunctionCount()295 uint64_t getMaximumFunctionCount() { return MaxFunctionCount; } 296 297 /// Factory method to create an indexed reader. 298 static std::error_code 299 create(std::string Path, std::unique_ptr<IndexedInstrProfReader> &Result); 300 }; 301 302 } // end namespace llvm 303 304 #endif 305